When performing distributed builds of machine learning packages, it
would be nice if builders without the required SIMD instructions can
be excluded as build nodes.
Since x86_64 has accumulated a large number of different instruction
set extensions, listing all possible extensions would be unwieldy.
AMD, Intel, Red Hat, and SUSE have recently defined four different
microarchitecture levels that are now part of the x86-64 psABI
supplement and will be used in glibc 2.33:
https://gitlab.com/x86-psABIs/x86-64-ABIhttps://lwn.net/Articles/844831/
This change uses libcpuid to detect CPU features and then uses them to
add the supported x86_64 levels to the additional system types. For
example on a Ryzen 3700X:
$ ~/aps/bin/nix -vv --version | grep "Additional system"
Additional system types: i686-linux, x86_64-v1-linux, x86_64-v2-linux, x86_64-v3-linux
Our use of boost::coroutine2 depends on -lboost_context,
which in turn depends on `-lboost_thread`, which in turn depends
on `-lboost_system`.
I suspect that this builds on nix only because of low-level hacks
like NIX_LDFLAGS.
This commit passes the proper linker flags, thus fixing bootstrap
builds on non-nix distributions like Ubuntu 16.04.
With these changes, I can build Nix on Ubuntu 16.04 using:
./bootstrap.sh
./configure --prefix=$HOME/editline-prefix \
--disable-doc-gen \
CXX=g++-7 \
--with-boost=$HOME/boost-prefix \
EDITLINE_CFLAGS=-I$HOME/editline-prefix/include \
EDITLINE_LIBS=-leditline \
LDFLAGS=-L$HOME/editline-prefix/lib
make
where
* g++-7 comes from gcc-7 from
https://launchpad.net/~ubuntu-toolchain-r/+archive/ubuntu/test,
* editline 1.14 from https://github.com/troglobit/editline/releases/tag/1.14.0
was installed into `$HOME/editline-prefix`
(because Ubuntu 16.04's `editline` is too old to have the function nix uses),
* boost 1.66 from
https://www.boost.org/doc/libs/1_66_0/more/getting_started/unix-variants.html
was installed into $HOME/boost-prefix (because Ubuntu 16.04 only has 1.58)
That was incorrect, because checking the dirent type already requires
a working compiler.
It had the effect that setting e.g. `: ${CFLAGS=""}` before `AC_PROG_CC`
as per `AC_PROG_CC`'s documentation would have no effect, because
`AC_STRUCT_DIRENT_D_TYPE` would automatically set CFLASGS.
(In a followup commit `: ${CFLAGS=""}` will be used, so it's important
to get this working first.)
And probably many other distributions.
Until now, ./configure would fail silently printing a warning
./configure: line 4621: AX_CXX_COMPILE_STDCXX_17: command not found
and then continuing, later failing with a C++ #error saying that some C++11
feature isn't supported (it didn't even get to the C++17 features).
This is because older distributions don't come with the
`AX_CXX_COMPILE_STDCXX_17` m4 macro.
This commit vendors that macro accordingly.
Now ./configure complains correctly:
configure: error: *** A compiler with support for C++17 language features is required.
On Ubuntu 16.04, ./configure completes if a newer compiler is used, e.g. with
gcc-7 from https://launchpad.net/~ubuntu-toolchain-r/+archive/ubuntu/test
using:
./bootstrap.sh
./configure CXX=g++-7 --disable-doc-gen --with-boost=$(nix-build --no-link '<nixpkgs>' -A boost.dev)
And probably other Linux distributions with long-term support releases.
Also update manual stating what version is needed;
I checked that 1.14 is the oldest version with which current nix compiles,
and added autoconf feature checks for some functions added in that release
that nix uses.
This turns previous compiler errors complaining about missing files
into proper ./configure time errors telling the user which version
of boost is required.
The unbalanced single-quotes cause many editor syntax highlighters
to interpret the rest of the file as a string literal, making it easier
to make syntax mistakes in absence of proper highlighting.
"configure.ac" is used to generate "configure", which is supposed to be
run with /bin/sh (as suggested by the shebang line of "configure"). As a
result it is a good idea to remove any /bin/sh-incompatible syntax from
configure.ac. Otherwise, systems that do not use Bash as their /bin/sh
are unable to run "configure" due to syntax errors.
Happily the failing tests should prevent anyone from using such a Nix
in situations where they expect sandboxing to be on,
which would otherwise be a risk.
* Look for both 'brotli' and 'bro' as external command,
since upstream has renamed it in newer versions.
If neither are found, current runtime behavior
is preserved: try to find 'bro' on PATH.
* Limit amount handed to BrotliEncoderCompressStream
to ensure interrupts are processed in a timely manner.
Testing shows negligible performance impact.
(Other compression sinks don't seem to require this)
This prevents builders from setting the S_ISUID or S_ISGID bits,
preventing users from using a nixbld* user to create a setuid/setgid
binary to interfere with subsequent builds under the same nixbld* uid.
This is based on aszlig's seccomp code
(47f587700d).
Reported by Linus Heckemann.
And add a 116 KiB ash shell from busybox to the release build. This
helps to make sandbox builds work out of the box on non-NixOS systems
and with diverted stores.
Build logs on cache.nixos.org are compressed using Brotli (since this
allows them to be decompressed automatically by Chrome and Firefox),
so it's handy if "nix log" can decompress them.
We're going to use libseccomp instead of creating the raw BPF program,
because we have different syscall numbers on different architectures.
Although our initial seccomp rules will be quite small it really doesn't
make sense to generate the raw BPF program because we need to duplicate
it and/or make branches on every single architecture we want to suuport.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Alpine seems to use this, and it results in a wrong
builtins.currentSystem. Big-endian ARM systems have triples starting
with armv6eb- or armv7eb-, so this doesn't change any systems that
already worked.
This reverts commit 80ebd60e7c. The
reason why we cleared CFLAGS/CXXFLAGS was because otherwise we get a
default value of -O2, which interferes with the defaults set in the
Makefile. (E.g. "make OPTIMIZE=0" should not pass -O2.)
